Upload
matthew-rice
View
212
Download
1
Embed Size (px)
Citation preview
The Problem with
Individual Risk
Beverly Rockhill, Ph.D.Department of Epidemiology
University of North Carolina, Chapel Hill
Primary Prevention of Disease
• Mass diseases and mass exposures require mass remedies--Geoffrey Rose
• Environmental, behavioral, and medical interventions should, and will, be targeted to each person’s genetic susceptibility--Muin Khoury
• Most identified chronic disease risk factors have only modest associations with disease—RRs 1.5-3.0
How to identify high-risk individuals?
• For most diseases, large majority of individuals will remain disease-free over considered time period, and “individual” risk estimates will tend to cluster at low end
• Bulk of disease cases will arise from mass of population with risk factor values ("individual risk") around average
How to identify high-risk individuals?
Boxplots of 14-year estimated risk of lung cancer, according to baseline
smoking status (NHS)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Estimated probability
Non cases Cases
Calibration vs. discriminatory accuracy – Calibration=goodness of fit; extent of bias
in model estimation. E.g., if average predicted risk for group of individuals is 0.10, and 10% of persons develop disease over time interval, model well-calibrated.
– Discrimination: ability to separate individuals with different outcomes.
Assessing accuracy of prediction at individual level
Risk factors as screening tools
Risk prediction tool must have large associated relative risk (>>20) comparing extremes of exposure or predicted risk in order to serve as useful screening tool at individual level
RRq1-5 = 2
Nondiseased Diseased
SENS = 8%5
RRq1-5 = 20
Nondiseased Diseased
SENS = 28%5
RRq1-5 = 200
Nondiseased Diseased
SENS = 56%5
Main points• Sensitivity and specificity, and resulting positive
predictive value, of most risk factors/risk models are poor. NPV higher, obviously, but key question is “how much is gained, given knowledge of risk factors, above and beyond knowledge of average risk/incidence in population?”
• Individuals concerned with these quantities, since they address question “what does this information mean for ME?”, rather than with good calibration, statistically significant predictors, etc.
Critical questions• Is there a systematic “rational” way that
individuals should act on “individual risk” information? Would a “rational”individual make a change in diet to lower 5-year risk of colon cancer from 15/10,000 to 8/10,000? Take a drug to lower risk of breast cancer from 2% to 1% in 5 years?
• Where does education end, and persuasion begin?
• Do communicators have full understanding of what they are communicating?
General conclusions
• As long as poor ability to single out small minority of individuals who will develop disease remains, a prevention strategy built upon idea of individual risk prediction and communication will have to affect many (i.e., persuade many to change or act) in order to prevent disease in a few
General conclusions
• “Mass remedies” needed; which are socially, ethically acceptable, and logically supportable?